In order to realize the online status monitoring and fault diagnosis of typical equipment during the operation of nuclear power plants, and to improve the timeliness of status monitoring and fault diagnosis of nuclear power plants, a technical framework of status monitoring and fault diagnosis based on artificial intelligence algorithms is proposed from the practical application level. Use common machine learning and deep learning technologies in the field of artificial intelligence to train different model algorithms through historical data of nuclear power plant operation, including normal data and abnormal data under different failure events. By comparing the test accuracy of different models, the most accurate condition monitoring model and fault diagnosis model are selected to apply to the online state prediction of nuclear power plants. When using real-time data to make online judgments on the state of nuclear power plants, first call the state monitoring model to monitor the abnormal working conditions of the nuclear power plant. Once the abnormal occurrence of the nuclear power plant is monitored, the fault diagnosis model is called to judge the abnormal event, and the fault of the nuclear power plant can be maintained and eliminated in time. Through the analysis and verification of the real data from a typical pump in a certain nuclear power plant in China, it is proved that the proposed calculation framework can be applied to the status monitoring and fault diagnosis of nuclear power plants. For nuclear O&M engineers who lack experience in developing artificial intelligence algorithms, this approach can reduce the difficulty of their development.